Stop Doing Monthly Reports for Analysts to Automate Finance Workflows Using AI with David Dors
In this episode of FP&A Unlocked, Paul Barnhurst sits down with David Dors to explore how artificial intelligence, particularly Claude, is transforming the future of FP&A. David shares his journey from banking and FP&A into finance transformation, where he now helps finance teams implement AI-powered workflows, automate repetitive processes, and rethink how work gets done in the Office of the CFO.
David Dors is a finance transformation consultant and AI educator specializing in helping finance teams adopt Claude, AI agents, and workflow automation. After beginning his career in banking and FP&A, David shifted his focus to helping finance professionals use AI to improve reporting, analysis, forecasting, and decision-making. Through workshops, training programs, and content creation, he teaches finance teams how to build AI-powered systems that save time and allow analysts to focus on higher-value strategic work.
Expect to Learn:
How AI is changing the future of FP&A
Practical ways to automate reporting and recurring finance tasks
The difference between Claude Chat, Claude Code, and AI agents
Why finance professionals should think of AI as a digital teammate
Key skills FP&A professionals need in an AI-driven world
Here are a few relevant quotes from the episode:
"Every forecast, variance analysis, or deck you build should help somebody make a different decision." – David Dors
"If nothing changed from the analysis you're providing, there's not much value in what you're doing every day." – David Dors
David explains how finance teams can use AI to eliminate repetitive work, improve efficiency, and spend more time supporting business decisions. He also shares his perspective on the future of FP&A, where professionals will increasingly manage AI agents that handle routine tasks and reporting. non-traditional organizations.
Follow Dave:
LinkedIn: https://www.linkedin.com/in/daviddors/
Website:https://vantagepointconsulting.com/
Earn Your CPE Credit For CPE credit, please go to earmarkcpe.com, listen to the episode, download the app, answer a few questions, and earn your CPE certification. To earn education credits for the FPAC Certificate, take the quiz on earmark and contact Paul Barnhurst for further details.
In Today's Episode
[00:00] – Trailer
[04:13] – Discovering Claude
[10:27] – AI Training for Finance Teams
[15:19] – Skills & AI Agents
[18:13] – AI Productivity Gains
[22:23] – Quick AI Wins in FP&A
[25:23] – AI Accuracy & Validation
[28:29] – Automating Reporting
[32:35] – The Future of FP&A
[39:01] – Excel Lessons Learned
[42:57] – Final Thoughts
Full Show Transcript:
Host: Paul Barnhurst (00:00):
Welcome to another episode of FP&A Unlocked, where finance meets strategy. I'm your host, Paul Barnhurst, the FP&A guy. Each week we bring you conversations and practical advice from thought leaders, industry experts, and practitioners who are reshaping the role of FP&A in today's business world. Together we'll uncover the strategies and experiences that separate good FP&A professionals from great ones, helping you elevate your careers and drive strategic impact. Today's guest is David Dors. David, welcome to the show.
Guest: David Dros (00:39):
Hey, thanks
Host: Paul Barnhurst (00:39):
Paul. Excited to have you. So let me give a little bit about David's background. He started in FP&A and today he is full-time working to implement Claude and FP&A software solutions for the office of the CFO. He started his career in banking. We won't hold it against them supporting private equity and wealth management clients. Then moved into FP&A before going full-time into finance transformation. He holds a degree from one of my Alma Mater, Arizona State University I where I did my MBA
Guest: David Dros (01:15):
Great school, number one in innovation.
Host: Paul Barnhurst (01:18):
It is a good school. I was there. I lived in Arizona from 2006 to 2016. No, 14, 20 14 I believe. So I was there for eight years.
Guest: David Dros (01:30):
Nice. Yeah, I've been here since I think around 2006 as well, but I've stayed.
Host: Paul Barnhurst (01:35):
Yeah, I was going to say, but you were probably a lot younger than I was then.
Guest: David Dros (01:38):
Yeah,
Host: Paul Barnhurst (01:40):
I'm old now. I'm interviewing people that I'm old enough to be their dad. That's, it's scary, but hey, what do you do? The last guy I interviewed is a business partner of mine on a project and one day we're sitting there chatting and I go, you realise I'm old enough to be your dad? Like I feel old right now. Alright, back fp a, everybody's like, just stop reminiscing. Paul, get to the point. This is the question we ask everybody. If I asked you to define what great FP&A looks like, what would you tell me?
Guest: David Dros (02:11):
Great. FP&A, I think it's definitely around finding the constraints of the business and then changing people's decisions on what they're doing day to day. So every forecast variance analysis deck you build, it's there to help somebody make a new decision or change the decision that they're going to make. So if nothing changed from the analysis you're providing, there's not much value in what you're doing every day. So I think definitely starting with, instead of starting with the data, starting with who's deciding what they believe in and what would make them believe something different? And sometimes it's just confirming their beliefs, just making sure that hey, we're going down the right path, but also it can be we need to make a change. And what would it take to actually convince 'em to make that change?
Host: Paul Barnhurst (03:05):
I really like how you said that if we're not helping people change decisions, make better decisions, and we're not really fulfilling the purpose of fp a,
Guest: David Dros (03:16):
There's always some constraint holding our business back. And if we're not looking for that every day, if we're not trying to change the decisions we make, we are just going to stay the same or even get worse. So if we're always trying to improve, it's finding that constraint that's holding us back and figuring out what would cost and what it would take to change that decision and change those constraints and remove 'em from the business till we find the next constraint.
Host: Paul Barnhurst (03:43):
And I really like how you mentioned constraints. I feel like I'm going back to my process improvement or kind of engineering classes and there usually is a bottleneck or constraint that's holding the business back. Not always, but I think it's a good way to look at it, looking for those constraints. That's definitely one way to think about it. So thank you for that answer. All right. So I know you are an early adopter of Claude. You've been spending a lot of time with Claude, so I'm going to start with how much time do you think you've put into Claude in the last two years?
Guest: David Dros (04:13):
Oh, I don't want to know, but I think it's open
Host: Paul Barnhurst (04:18):
Now. Your friend, do you spend more time with her than anyone else?
Guest: David Dros (04:21):
Yeah, it's a replace Excel for being that thing that's always open 24 7. It's like Claude's just always open when I work, that's the first thing I reach for. So yeah, it's got to be eight plus hours a day since 2025 at this point, beginning of 2025. It's a good chunk of my time now.
Host: Paul Barnhurst (04:43):
Do you have a name for it? Do you consider it a he a she and it. I always find that interesting how different people think of Claude.
Guest: David Dros (04:51):
I've been just Claude. But on April 1st they released these buddies. They were like Pokemon buddies or something. You have an egg that hatches and I got one that was a cactus, so I really liked that one. Vanilla took it away. They took it away after April 1st, so it was nice to have my little friend down there. No, he's gone.
Host: Paul Barnhurst (05:12):
That's funny. I didn't see that. My April 1st was I was launching a beard brand,
Guest: David Dros (05:19):
A beard bomb.
Host: Paul Barnhurst (05:21):
I had beard bal, beard oil and beard butter.
Guest: David Dros (05:23):
I would buy, I'd be one of the first buyers.
Host: Paul Barnhurst (05:26):
I've had a few people tell me that you never know. I might just launch it one day, put it on, I already have a page on the website I just got to to get the product.
Guest: David Dros (05:34):
Yeah. Have Claude help you build the
Host: Paul Barnhurst (05:37):
Business? Oh, it helped me build the website and the images. I think I use a little bit of chat. GPTI still find, and I know we're digressing for a minute, but I find when I want just a jpeg, a PNG image, I still prefer chat GPT to Claude. Claude's got a lot better there. I know most people prefer Gemini, but it's like how many different tools do you want to use? I pay for three. I'm like, I want to pay for a fourth.
Guest: David Dros (06:00):
Yeah, it is tough to find the balance between going deep into one tool or using what's best from every tool. But they just released the claw design and I messed around with that and that can make board decks, it can make websites, so it's getting better there. I think it's improving.
Host: Paul Barnhurst (06:20):
I have not played with Claude Design. I didn't see that release. I know they just released 4.7, so I'll have to take a look at Claude Design. But yeah, all this keeps getting better. Sometimes it's a little scary, but it's exciting. What made you lean into Claude? When were you convinced that, hey, the tool I really want to spend my time in? Was it because of work or how did that come about that you made the decision? I'm going to go deep into Claude.
Guest: David Dros (06:44):
Yeah, at first I was definitely, like you were saying, switching between these tools, like using Claude for some things, chat, GBT, chat for others, and then they released Claude Code and we're finance guys, we don't code, but I was like, I wonder what this is. And I looked into it and I saw that the workspace was your folders and what it did for you was create files or edit files. And that was like, that's exactly what I do. I work inside file folders and I go and I edit Excel files and PowerPoints. When I saw that I leaned into a clot, I was like, oh, they're leaning towards this worker audience. It's not chat where I'm just going and asking for a formula for Power Query going in there and it's doing work for me. So I leaned in then and then they really convinced me. They just kept policing things, CLA in Excel, CLA in Word. So they're all just leaning towards helping you do work better. So yeah, it's work helping me do work better.
Host: Paul Barnhurst (07:49):
So Claude Code was kind of really the first one that you saw because it had that workspace and it could work on files and said there must be something here. How can I use it?
Guest: David Dros (07:57):
Exactly. I knew AI was working for engineers, how can I make it work for finance?
Host: Paul Barnhurst (08:04):
Yeah, there's a lot we can learn. I interviewed a guest who said, look, engineers, developers are 18 months ahead of us. Learn how they're using Claude and copy a lot of, or whatever tool, but learn how they're using AI and copy a lot of it.
Guest: David Dros (08:18):
Yeah, yeah, exactly that. And they're a little more blessed than us. They have really great version control, which we don't have in Excel, which I'm very jealous of using tools like dang, I wish Excel would do that. We can collaborate so easily across engineers,
Host: Paul Barnhurst (08:36):
But it's funny, I was talking to somebody that's building a software where one of the things they're talking about is we're trying to manage the version control and they still output stuff to Excel is an interesting approach with using ai. And so I think we'll see more and more solutions to get better and better on that front. And so it'll be fascinating to watch. There's some modern spreadsheets that do some of these things really, really well. But as I've always said, you're competing against the 800 pound gorilla and the most complete software in the world as far as spreadsheets go is Excel. And in my mind there's probably not a close second of all the different things it can do. Google Sheets can do 90%, it can do some things way better. It's probably the next closest, but they just have invested so much in the whole Microsoft ecosystem. It's a really hard disconnect.
Guest: David Dros (09:25):
It's been where we've worked. Everything we do. If the software we have can't do something, we reach for Excel. So it is hard to beat that.
Host: Paul Barnhurst (09:35):
It's a fabulous tool. I mean I love Excel and AI hasn't changed that. I can see areas where it's changing how we work with it when we should work with it for sure. And I continue to see that, but there's a reason chat, GPT, Claude and Copilot all have add-ins to Excel.
Guest: David Dros (09:52):
Exactly. They know they can't compete, but they're like how do we make this better, make the experience better, which I
Host: Paul Barnhurst (09:59):
Correct. Plus they're getting all that data, which maybe at some point they do compete with that, but for now they recognise that having all that information and building something that works well with them is a great way to go. Alright. I know you've started your own business now you're doing training for finance teams. Why is this so important? Why did you see that was a need and that you felt, hey, I'm going to leave my steady day job to start training people?
Guest: David Dros (10:27):
Yeah, I think it was just that aha moment I had. I want everyone in finance to feel that and I feel like once you get there it just doesn't stop. I have a lot of friends that use these tools now. They use cowork or code every day and I'll ask 'em like what problems do you have now that you're using this? And it's basically gone away. They're like, I don't have any problems. I just go and I build my own solutions to problems and you're not looking for software to solve these problems now you have kind of the whole world open to you outside of Excel now. So that's why I've been focused on the training of just how do we get people to that aha moment as quick as possible.
(11:12):
It is confusing the technology. There's a lot of different things where we start talking about skills and CPS and projects. If you don't know what any of that is, it's like I'm learning to be a software engineer or something. But once you get in and just start using it for your actual work and you can see how it starts saving you time and lets you focus on different things that you'd much rather spend your time doing. It's a great tool and I think more and more finance will start using this as a platform where they go and do work.
Host: Paul Barnhurst (11:44):
Let's break down kind of the different things in a simplistic term, how you would describe 'em. I think this is interesting. So let's start with we have Claude where I can just go to the browser and open it up. So I'll call that kind of Claude chat. We have Claude Cowork, we have Claude Code, we can download it to our desktop, we can download it to our phone. Now we have Claude Design. So let's start there. I mean what's kind of the difference? How should people be thinking about this whole ecosystem almost this operating systems so to speak, that Claude's built? How do you think about it?
Guest: David Dros (12:23):
I don't reach for chat too often now just because of everything I've built for myself and the other tools. But chat I feel like is just the same thing. We've been kind of using it forward quick questions like I want help with a formula or I just want some quick information. That's how I've been using it. And I feel like once you get into coworker code and you start building out your own personal operating system there, that's where you stop using the chat tools and you switch to either coworker code, they're very similar tools. I prefer code just because that's what I've been using, but Cowork just has kind of the nice UI in front of it so it looks pretty, it's like what we're used to using. And when you get into these tools, you're basically building connections to all the data sources that you already use, whether that's your ERPA data warehouse, even your email. And then you can have co-work do the work you would normally do, whether that's analysing data, cleaning it up, transformation, and you can build out what they call skills, which is just, they're all markdown files, which is just very much, it's a text file that explains your workflow to Claude. If you do something a specific way every time you turn that into a skill and then Claude can go and do that every single time.
Host: Paul Barnhurst (13:57):
So I want to break that down a little further. So we've talked about kind of code chat. I mean chat is really where you go ask questions. If you're doing the serious work, it's do I want to try to do it in co-work or code but also isn't also designed where it will access your computer and do tasks for you? Does code do that as well?
Guest: David Dros (14:16):
Yeah, exactly. So code can do that as well. Code actually opens in the terminal, which is fancy engineering. It's not that scary once you get used to it, it takes a few minutes. I watched a YouTube video on how to get in there, but all it does is it opens inside one of your folders. So it can create a file, it can create an Excel file for you, it could write code for you and that happens all on your computer. So both of them are using your computer.
Host: Paul Barnhurst (14:47):
So both of them could do that. It could go to a website, it could open up an application if needed in code, just want to make it clear for the audience that you could pretty much almost anything you can do in cowork, you can do in code. A code also has kind of the whole code side of it.
Guest: David Dros (15:02):
Code has more features that some things that haven't come to cowork yet. Code actually does a little better still with things like memory where it'll start to remember your workflows or your preferences and it'll store that memory. So next time you go and do it, it remembers.
Host: Paul Barnhurst (15:19):
Alright, so now let's talk. So we have these three. The next thing you mentioned is skills, which you said markdown files. Way to think of 'em. The way I describe 'em is instructions, right? So you're giving it instructions. It really provides a purpose for good documentation. Usually something everybody was terrible at. Now you want to document it because if you document it well you can give that to ai. So how does someone get started with the skill? It feels like I got to give it all this instruction. Isn't it a lot of work to write all the documentation?
Guest: David Dros (15:49):
Yeah, it does take some time. Once you do your first one, you'll start to understand the patterns of you're not trying to give it step-by-step instructions. You're really trying to give it the tools it needs to go and do work that you would do yourself. So if something you're doing constantly is exporting a specific report, you can give it access to that report and tell it how to actually go and pick the right timeframe or the right setting so it brings that report back accurately. So yeah, it is like instructions, but also you're just giving it how it's done and some of the why of what we're doing and that lets it act more autonomously and get through and give you, it surprises you with, once you start giving it different tools to access, it can really go and find creative solutions to problems you didn't even know you had.
Host: Paul Barnhurst (16:52):
Okay, so we have these skills, we have the markdown files, you give it context, you give it why, you give it some instruction but not necessarily a step-by-step. Give it that freedom. How big of a difference does it make using a skill versus not using one? When you're doing something inlaws, you give it a prompt, ask it to do the same thing. Are you finding it's really a key difference?
Guest: David Dros (17:13):
Where skills really start to help is when you're doing something every day, every week or every month. And so you're not constantly re-explaining the same prompt. You're giving it all this context that it needs to do the work correctly and you can put it inside the skill. And the skill is not always just the markdown file like we talked about. You can give it examples, you can give it assets, things to go and see how we did this previously and it can go and rebuild it exactly the way we did it before. So something like management reporting, if we're always using the same format, we don't want to change the format, we don't want to change it. We give it as a asset or a reference for it to go back and recreate that. Exactly.
Host: Paul Barnhurst (17:59):
The model after it.
Guest: David Dros (18:00):
Yeah, exactly. So you always get the same output. Makes
Host: Paul Barnhurst (18:03):
A lot of sense. I'm curious, how much time do you think you're saving on average a week? Let's talk, what were you saving from Claude and then maybe also in your personal life?
Guest: David Dros (18:13):
Yeah, no, honestly it's so hard to say because every time you cut something out, you always find something that fills it. So you start cutting some of the repetitive work that doesn't really require my judgement . It's like pulling data when we're cleaning it the same way. Reformatting that can cut five, 10 hours every month on just one process that you do every month. But something always finds the way to fill in that time gap and I'm not worked any less. So it's like you start working towards things you like to do more and things that you find more interesting.
Host: Paul Barnhurst (18:52):
Yeah, it's basically allowing you to focus more on what you want versus oh now I work six hours a day.
Guest: David Dros (18:59):
Yeah, it does let you focus. That's one of the things that I found really helpful is instead of switching between all these tabs, going between Excel and different browsers and email, I can work in one workspace and that helps me just focus on what needs to get done and I start de in the other work to agents that go and work for me.
Host: Paul Barnhurst (19:22):
Okay, got it. Next you recently you did, I think it was a post on LinkedIn, you shared the Claude is the way you see it, it's not really a software you deploy, you shouldn't think of it as a software, but you should think of it as someone you onboard as a new hire. Talk to that, talk to why you think about it that way and what's the difference when you treat it like a new hire versus a software?
Guest: David Dros (19:44):
Yeah, so when you go and use Claude, you don't have just one, you might have a different one for each workflow you do. It could be one for data analysis.
Host: Paul Barnhurst (19:54):
And when you say one, are you saying one project, one skill? What do you mean by one?
Guest: David Dros (19:58):
Yeah, so the way I think about agents or cowork is the folder is the agent. So I will have a different folder for different types of work.
Host: Paul Barnhurst (20:09):
Got it.
Guest: David Dros (20:09):
And inside that folder I have different skills that are specifically tuned to what the work I do inside that folder and it might have different data access inside those folders.
Host: Paul Barnhurst (20:21):
And do you have to also load those skills into cla? So do you just keep 'em in case you need to change 'em in the folder, but they're also loaded into Claude or how's that work?
Guest: David Dros (20:30):
Yeah, so I use Claude code, so most of everything I have is saved locally on my computer, but inside cowork you can save them to the cowork app and so now they're saved in the cloud and so you can get 'em wherever you need them, which is really nice. But yeah, when we talk about your onboarding an agent, each one needs different contexts, different instructions, different data access, what we talked about with skills and connectors and they get better over time where software, you set it up, you train people and it works the same way basically forever. There might be some updates but it's basically the same. But with your agent they're constantly improving. We already talked about leasing all these features every day and you constantly have to work and re-update these agents, make sure they work correctly still, and when you're no longer managing one agent, you're managing 10, it's a completely different structure. They still need the same things the team needs where it's like a clear scope, good data, the right authority and somebody to actually own these outputs. So that's why I think it's more of building this environment for a team of agents to go work in instead of just a software that you turn on one day.
Host: Paul Barnhurst (21:55):
Got it. So from the new hire perspective, it sounds like it's really about giving it the context and training it. That's kind of a similarity.
Guest: David Dros (22:04):
Yeah, exactly. You're teaching it your business model, the quirks about your business, why something gets done a certain way and what you would learn when you're starting in a role. Your job is just now to teach Claude the same things that you learned when you onboarded and when you took on different responsibilities.
Host: Paul Barnhurst (22:23):
So what is the low hanging fruit that you think FP&A people shouldn't be doing anymore because of ai? Whether it's Claude chat, GPT, copilot, what's one or two low hanging fruit that you would say, Hey, if you're doing it this way you should be rethinking how you're doing it?
Guest: David Dros (22:38):
Definitely around accessing data where the way we currently access data is you go into your ERP and you click export a report and then you go and you configure it and you say, I'm doing this time period, but your agent can just go and get all that data for you without going through a UI and taking all that time. It just goes and it runs a connection that gets that data right back for you. So just those things like that where they might take five, 10 minutes, but over the course of a month a year, it adds up to hundreds of hours where why am I doing this work if it doesn't require my judgement , I'm just exporting the same report every time I think it should go to agents.
Host: Paul Barnhurst (23:26):
Got it. So the first area, the lowest hanging fruit is exporting data, the manual exporting,
Guest: David Dros (23:31):
Just giving it access there. Then a lot of the transformation work we do is it's the same every month. If we're doing the same calculations, that logic doesn't change that can go to your agent next you're just rebuilding your workflows but in a way that an agent can do them.
Host: Paul Barnhurst (23:51):
Got it. And if somebody's listening to the show and they're trying to figure out where to start, do they start with some CLO training? Do they start with trying to automate something From your opinion? Where would you recommend they start? This is somebody, let's assume they've done prompting in AI but they haven't automated any processes, they haven't built an agent, they don't know what a skill is, they've heard the terms, but they haven't done anything. What would you recommend is that place to start?
Guest: David Dros (24:21):
Yeah, when I think about it for a team, like a team of analysts or a finance team, I think it should start before the tool, before you even get into Claude, just making sure everybody has access or any sort of friction before actually going and getting to that aha moment I was talking about. It's just going to stop people from using it. So if they don't know what data they can use in there or what cloud can actually access, I would start there and make sure everything's explicit where everybody knows what they can and cannot do with it. That really just helps for once you go in there and you start using the tool. If you have access to everything you need and it can do the work, you can go and build the solutions your own way, whatever way you see fit, but it's really about having that access in the beginning and just figuring out that governance layer.
Host: Paul Barnhurst (25:23):
What are the challenges? We hear a lot about hey, AI hallucinates, it doesn't do things the same. What are the things you need to be watching for? I'm assuming whether you have an agent doing it, whether you are prompting throughout working with it and it's a back and forth process, there needs to be this human in the loop. So how do you think about all that? What are the things people should be watching for? Because just like a human, we know this can make mistakes,
Guest: David Dros (25:51):
So we want to have our processes be as deterministic as possible before they get to the agent. If the data can be made up by ai, then everything after that's going to be terrible. So making sure we have access to data that is deterministic and it's going to be accurate a hundred percent of the time, that should be your baseline. And then now once you're going in and doing the actual work and transforming data for finance, we will almost always need a human in the loop to go and validate those outputs. But it's giving AI a way to actually check its work and make sure it actually is accurate. Just like how we do with checks, like check sums and you're always validating all the data you want to give agents the same tools you have to make sure we can go and check and make sure everything we did is accurate. Everything ties, everything checks out. So giving agents a way to go and evaluate their work is a really good way for you to not be fixing a lot of small problems and you're fixing more root cause problems instead of little one-offs.
Host: Paul Barnhurst (27:07):
And elaborate on that. Is that something in the instructions? Is that reference sheets, is that giving them access to another tool where they're supposed to check the number? What do you mean by that? So I built my agent, I run it and the numbers come back, let's just say quite a bit off. How do I get it to check most of it itself? Is it writing in your instructions, check two or three times specifically look at these areas, giving it access to other data? What are some of those things that reduce that variability in the errors beyond, hey, I gave it a skill, I gave it the data, I gave it an example
Guest: David Dros (27:43):
With like say we go and we have a build the Excel report for us. So just having it billed in the same checks we would use where we can go and make sure all the data ties out to the different source data and giving them the checks we can do at the skill level where we're giving it the checks that we would normally do when we go and look over and review our work and telling it, Hey, go through and make sure you validate these three things. Or there's also the option of having more of a review agent where there's another agent that reviews work specifically. And so you have one agent that does work, another one that reviews it all before it comes to you.
Host: Paul Barnhurst (28:29):
So let's walk through an example we talked about, let's take reporting. You mentioned we need to reduce frictions, making sure you have access to everything, you understand the tools and then you also talk about, hey, you're going to need to write the skill. But I would imagine another key area here is process. You need to be able to document, outline what that process should work like. So let's walk through kind of how all of that look. So we're going to automate our monthly p and l reporting as an example, the process and all that. Walk me through that.
Guest: David Dros (29:03):
So the first thing I would do is figure out what your actual process is. We know the process we go through every month and do, but it's a lot different when you start and try and explain this to an agent, build a skill, build the tools that can go and do this for us. So I would definitely go and use one of these dictation tools where you can use voice and just go and talk to cloud for a minute and explain your process, tell it what the inputs are, what transformations you do, and then what the outputs are and what they should look like. And that really gives us a good starting point to go and start building these automations. And then once you have that mapped out, you can see where biggest bottlenecks are, what takes you the most time. Sometimes it could be really easy to go and pull the data yourself, it might take you a minute, but what really takes a lot of time is the formatting and getting everything into PowerPoint decks. And so that might be the process you want to start with first. And so building out the skills that your agent would need, everything they would need to do to do it the same way you would do. And we can create that as a skill, give it our processes, give it access to the data sources it needs, and it can go and do the transformations for us if we need or it can go and do the formatting and building out the reports.
Host: Paul Barnhurst (30:30):
Got it. No, that's helpful. Thank you. What's the coolest thing? Talk more personal life you've built with Claude code. Something you've vibe coded.
Guest: David Dros (30:41):
Oh, something I've vibe coded
Host: Paul Barnhurst (30:43):
Or just a process you've done. My last guy, I'll give example,
(30:47):
He goes, the only area my wife and I ever argue over is what mill we're going to do. So they use Claude, he goes, I never want to do a mill plan again in my life. And it built an app that helps him do all their meal planning. He used an open source and he built it all. I mean it could be totally different, but that's an example of kind of a fun one I had. He is like, now we just use the tool to figure out what our meal plan is based on what we have, what's easy to get, what will be healthy and away we go.
Guest: David Dros (31:14):
Yeah, no, I did that where I vibe coded a Apple app actually on my phone where I like to track if I have any habits I like to do working out every day or eating good foods and not sticking to my habits. So I had a vibe code, an app for me that was, I kind of tried to gamify it where it was every time I did my habit I would log in the app and it would level me up. I started getting experience and my little guy started getting higher levels, he was getting better.
Host: Paul Barnhurst (31:58):
Now did he get buffer as he went stronger?
Guest: David Dros (32:02):
Yeah, so that was one that I did and I've never used, I forget what the software is, but it used the software for me inside to create a Apple app and I've never done anything like that. So that was really awesome that I'm able to do that and now it's on my phone.
Host: Paul Barnhurst (32:19):
That's cool. Yeah, I'm hearing just amazing stories of what people are doing. If you didn't see it, you'd think we live in a sci-fi world sometimes with the stuff people are able to do. Alright, so I want to ask this, what's fp a going to look like in five years?
Guest: David Dros (32:35):
Man, it's so hard to even know what two years
Host: Paul Barnhurst (32:37):
And I know we're guessing. I mean if you are right, I'm coming back to you and we're going to Vegas.
Guest: David Dros (32:42):
Yeah, exactly. The way I see it is everybody's going to be managing a team of their own agents. So everybody's going to move up a level. If we're the analysts now we're the analysts that manages 10 other agents that did the work we used to do. So when you're talking about how are we going to train the next generation, it's going to be really tough when agents are going to handle a lot of the work we do today. It's really going to change just the way a lot of businesses operate. If you're already seeing, there was somebody that had a billion dollar business and it was just him and his brother and they ran the whole business using AI agents. So the way FP&A is going to change, I think we're going to be embedded into a lot of different processes, helping make decisions across the business where agents are going to handle a lot of the standard reporting. We do building forecasts and we'll be off trying to help the business.
Host: Paul Barnhurst (33:45):
Got it. Alright. If you had to go back before Gen ai, what's the one thing you just couldn't do anymore? You couldn't go back to the way it was before.
Guest: David Dros (33:53):
It's really just the work that I already hated doing where when you're doing the same thing every single month for years, it's like nothing changes about it. You do the same exact process. I would never want to go back to that where I didn't know how to automate some of this work and I was just doing the same thing every month.
Host: Paul Barnhurst (34:18):
Fair enough. So it's really, it's the repetitive tasks, you just couldn't do that again.
Guest: David Dros (34:23):
Yeah, exactly. The things where I wasn't adding any value, I was just doing the same thing every single tonne.
Host: Paul Barnhurst (34:29):
That's my career. I don't know if I ever added value.
Guest: David Dros (34:32):
Yeah, you
Host: Paul Barnhurst (34:33):
Did. I couldn't admit that.
Guest: David Dros (34:35):
I think everybody learns so much from this. Just being able to hear other people's experiences and it gives them ideas on what they can go and do.
Host: Paul Barnhurst (34:44):
That's one of the reasons I'm so excited about doing the AI series. There'll be a few other episodes intermixed, but you're the fourth person. I've done a heavy, Hey, how are you using ai? Walk me through things, let's talk about examples. And I'm learning a lot. I am already thinking I need to do this and I need to do that. And it's exciting. So I'm hoping others get the same out of it in their day-to-day work. I really think spending the 10 hours, let's say I do 10 episodes, probably even five hours, listen to about double speed, you'll get some great ideas. But the key is as you know, you have to do listen. Watching YouTube, you can do that all you want, but you just have to jump in. That's the thing. I'm realising more and more, and I've started to build a couple skills. I've started to play with it a little bit and I know I just need to jump in more. It's just as simple as that.
Guest: David Dros (35:35):
Once you dive in, there's no going back.
Host: Paul Barnhurst (35:37):
And I'm starting to realise that more and more as I, I'll give an example. I'm building a deck, I'm doing a keynote speech for an event and I've probably spent three, four hours in PowerPoint. Okay, I've got this section done, here's what I'm trying to accomplish. Give me ideas, tell me how I'm thinking about. And sometimes I'll have it redo the slide. I may still edit, I'll edit 'em after. And I know I always like the look and feel and everything isn't quite like I want, but it's incredible how much I can use it as a thought partner. Ignoring the agents, ignoring the automation, just the thought partner part to augment what you're doing is incredible. Then you add all the efficiency gains If you go, which is where I need to go that next step and how do I automate my workflows? We have a CFO on who said, I look at AI as my thought leader.
(36:24):
It's kind of a lonely position. There's not many people I can go have certain discussions with in the company, but I can go use AI as my thought partner. And so it's just so versatile. Kind of like you could do anything in Excel. There's a similar with ai, it's kind of that new world that there's the tool that is going to be as ubiquitous, right? It's gen ai, whether it's Claude or Chat GPT or copilot. We'll see which one it is. Everybody has opinions. I have mine, but I think you're kind looking at that, the modern day and more powerful costs more areas, but kind spreadsheet in some ways, if that makes sense. But that's a little bit of how I think of it.
Guest: David Dros (37:02):
Oh yeah. The options are unlimited. You can go build whatever you want to build now. You're not forced into what somebody else wants to build for you and fitting cleanly into what they built. Pretty
Host: Paul Barnhurst (37:15):
Wild.
Guest: David Dros (37:15):
Yeah. Customised as you want it to be.
Host: Paul Barnhurst (37:18):
Yeah. I have so many things I need a three months off to just be a AI person anyway. Alright. What's the number one technical skill that FP a professionals should master today?
Guest: David Dros (37:30):
I don't think it's Excel anymore, but I think it's understanding what Excel does for you. The operations underneath what happens when you do a lookup or when it creates a pivot table. Because AI can do so much for you now where if you can direct it at the most base level of what it's doing, we're cutting out these UIs, we're not going in and going through our ERPs, nice screen and clicking download p and l for January. AI can do that for us and it can do all the transformations that we want it to do as long as we understand what it actually needs to do, not how to do it in Excel, but how do you do it at the most base level.
Host: Paul Barnhurst (38:14):
Got it. Alright. What's the number one soft skill we should master?
Guest: David Dros (38:19):
I think clear writing because growing up you were here in Arizona, our schools, they teach us how to write these long essays. You're always trying to hit some character limit, get past a certain amount of words, you're stuffing with unnecessary words and then you get into finance and nobody cares that you know how to use all these nice words. They just want to know what decision do I need to make? How do I make it just be clear, be simple,
Host: Paul Barnhurst (38:50):
Concise, communication, written, verbal, whatever it might be.
Guest: David Dros (38:54):
Yeah. I really wish schools would focus on that side of writing more because I feel like you don't get that.
Host: Paul Barnhurst (39:01):
All right. I know you've used Excel a lot. What Excel mistake taught you the most in your career?
Guest: David Dros (39:08):
Trying to make the most the crazy nested lookups index matches that have all these dependencies and that you never know when something breaks. I would try and fit everything into one formula just because I could. And then I realised I don't even know what's going on in here. And when you send something out that's just completely off because of a formula, you feel so stupid and you're just like, why did I do it this way? Break it out cleanly. Show all the steps. And
Host: Paul Barnhurst (39:43):
I had a guest once who said, open your model a week after you built it. If you don't understand the formulas, it's too complicated. And I've opened files where I'm like, what was I doing here? And you have to pick it apart. That's a sign that you did a poor job. And we all do it because at the end of the day, design, design design is critical to excel.
Guest: David Dros (40:06):
Yeah, that's why I love Power Query so much where you really just listed out all your steps and now go into using AI so much where it doesn't really work well with Power Query. You're back to the basics. So that's why I've been trying to relearn. Yeah,
Host: Paul Barnhurst (40:23):
They'll figure out how to make it easy. Bring AI into the Power Query interface at some point or they'll figure some other solution, but it's a crazy world. Alright, this is the Get to Know You section as we wrap up, if I asked you where you would go, if you could go anywhere in the world tomorrow for a vacation to go see a place you haven't been or maybe even somewhere you've been, where are you going? I've
Guest: David Dros (40:44):
Going to Japan.
Host: Paul Barnhurst (40:46):
Why?
Guest: David Dros (40:46):
I think the food is amazing. Everything I've had over here, at least that's Japanese. I think their country is beautiful, the culture is just amazing. They just get so obsessed with different things and I love how obsessed they are about mastering one thing. And if you ever read Shoe Dog from Phil Knight, who was the,
Host: Paul Barnhurst (41:09):
Of course. Yeah, yeah, yeah. I read that
Guest: David Dros (41:10):
From Phil Knight. He's the founder of Nike. He just had this life-changing experience when he went to Japan, so going in October. Awesome.
Host: Paul Barnhurst (41:19):
Well let me know how it goes. Excited for you. Alright. You get to spend one day one day only with any person in the world. So somebody alive, who are you spending the day with? And obviously outside of me of course.
Guest: David Dros (41:33):
Alright. All right. Now I got to think about it. My first option's gone.
Host: Paul Barnhurst (41:37):
Yeah, whatever.
Guest: David Dros (41:39):
If I can do two, a little combo, my wife and son, I already spend every day with them, but
Host: Paul Barnhurst (41:46):
Smart answer, she listens, she'll like it. And I know it's legit, I get it.
Guest: David Dros (41:50):
But no, it really is legit. They're why I do everything. So just another day with them, I'll take it. I'll take every day I can with them.
Host: Paul Barnhurst (42:00):
I love it. Alright, if someone wants to learn more about the services you offer, this is your time to kind of plug what you're doing as far as services. Share a little bit about that and how they can contact you.
Guest: David Dros (42:11):
Yeah, so LinkedIn's probably the best way to reach out to me. David dors. I share a bunch of information on Claude, how to set that up for your finance team hosts a lot of workshops where we go in and we build with cloud, cowork or cloud code. And then I have a lot of longer how-to articles on how to get this set up. So if you're building anything cool with Claude or in finance, just message me on LinkedIn and I want to know what you're building.
Host: Paul Barnhurst (42:41):
All right. Awesome. Well, thank you so much for joining me, Dave. I appreciate it. I hope you enjoy the rest of your day and excited for the audience to get to listen to this. So thank you for coming on the show.
Guest: David Dros (42:51):
Thanks, Paul. It was great to be here and it was great conversation about where we're going with finance.
Host: Paul Barnhurst (42:57):
Yeah, it's an exciting time. So I'll leave with these party words. If you're not using ai, just get on the boat, just start, even if it's simple, I can't say by any means. I'm an expert, I'm behind a lot of people, but I'm continuing to jump in and that's all you can do.
Guest: David Dros (43:11):
You just got to get in there and start using it.
Host: Paul Barnhurst (43:13):
Alrighty, well that's a wrap. Thank you everyone. That's it for today's episode of FP&A Unlocked. If you enjoy FP&A unlocked, please take a moment to leave a five-star rating and review. It's the best way to support the FP&A guy and help more FP&A professionals discover the show. Remember, you can earn CPE credit for this episode by visiting earmarkcpe.com. Downloading the app and completing the quiz. If you need continuing education credits for the FPAC certification, complete the quiz and reach out to me directly. Thanks for listening. I'm Paul Barnhurst, the FP&A guy, and I'll see you next time.